ESTIMATE statement: age-adjusted rates using Poisson models in PROC GENMOD

RonaldB23 — Sat, 05 Apr 2014 22:32:56 GMT

Hello everyone and thanks in advance for helping me out with my problem (or taking the time to read my post),

I am trying to calculate age-adjusted incidence rates (not ratios) for a medical procedure for each calendar year in my data set (2000-2012). See the code pasted below. For some reason when I run this, I get all zeros back for my incidence rates specified in the ESTIMATE statements. I tried exponentiating the beta coefficients myself and got the same problem (results *10^-11). Again, results included below.

Now, I know I shouldn't do this, but when I add age agesq ageneg1 to the CLASS statement my estimates seem to make sense. Still - I don't trust them since I want age modeled continuously, not categorically.

My data is in the form of aggregated count and person time (one observation for each year of age, for each calendar year). Please let me know if working in this aggregated format is the problem, because I can obtain more detailed data.

Thanks for your time! Sorry if I'm missing something very obvious. I've been searching through the documentation for SAS procedures and nothing helped me to see the error of my ways.

Here's the code I'm trying to run:

PROC GENMOD DATA=work.CountData_AgeCyear

CLASS CalYr;

MODEL TotalCount = age agesq ageneg1 CalYr

/ DIST=poi LINK=log OFFSET=log_pt SCALE=deviance;

ESTIMATE "IR: 2000" int 1 calyr 1 0 0 0 0 0 0 0 0 0 0 0 0;

ESTIMATE "IR: 2001" int 1 calyr 0 1 0 0 0 0 0 0 0 0 0 0 0;

ESTIMATE "IR: 2002" int 1 calyr 0 0 1 0 0 0 0 0 0 0 0 0 0;

ESTIMATE "IR: 2003" int 1 calyr 0 0 0 1 0 0 0 0 0 0 0 0 0;

ESTIMATE "IR: 2004" int 1 calyr 0 0 0 0 1 0 0 0 0 0 0 0 0;

ESTIMATE "IR: 2005" int 1 calyr 0 0 0 0 0 1 0 0 0 0 0 0 0;

ESTIMATE "IR: 2006" int 1 calyr 0 0 0 0 0 0 1 0 0 0 0 0 0;

ESTIMATE "IR: 2007" int 1 calyr 0 0 0 0 0 0 0 1 0 0 0 0 0;

ESTIMATE "IR: 2008" int 1 calyr 0 0 0 0 0 0 0 0 1 0 0 0 0;

ESTIMATE "IR: 2009" int 1 calyr 0 0 0 0 0 0 0 0 0 1 0 0 0;

ESTIMATE "IR: 2010" int 1 calyr 0 0 0 0 0 0 0 0 0 0 1 0 0;

ESTIMATE "IR: 2011" int 1 calyr 0 0 0 0 0 0 0 0 0 0 0 1 0;

ESTIMATE "IR: 2012" int 1 calyr 0 0 0 0 0 0 0 0 0 0 0 0 1;

RUN;

Here are the parameter estimates I get (all 0's are output from my ESTIMATE statements [not displayed]):

Parameter DF Estimate Error Limits Chi-Square Pr > ChiSq

Intercept 1 -24.3139 0.3298 -24.9603 -23.6675 5434.65 <.0001

AGE 1 0.3573 0.0068 0.3439 0.3707 2737.97 <.0001

agesq 1 -0.0018 0.0000 -0.0019 -0.0017 1798.56 <.0001

ageneg1 1 179.2731 4.6522 170.1550 188.3913 1484.96 <.0001

CalYr 2000 1 -0.0399 0.0567 -0.1511 0.0712 0.50 0.4813

CalYr 2001 1 -0.0313 0.0431 -0.1158 0.0533 0.52 0.4688

CalYr 2002 1 0.0528 0.0360 -0.0179 0.1235 2.15 0.1430

CalYr 2003 1 -0.0556 0.0316 -0.1174 0.0063 3.10 0.0783

CalYr 2004 1 0.0254 0.0282 -0.0299 0.0808 0.81 0.3678

CalYr 2005 1 0.1314 0.0265 0.0794 0.1833 24.53 <.0001

CalYr 2006 1 0.3067 0.0256 0.2565 0.3569 143.36 <.0001

CalYr 2007 1 0.3010 0.0254 0.2512 0.3509 140.00 <.0001

CalYr 2008 1 0.3772 0.0224 0.3333 0.4211 283.17 <.0001

CalYr 2009 1 0.3541 0.0224 0.3101 0.3980 249.30 <.0001

CalYr 2010 1 0.3451 0.0223 0.3013 0.3889 238.80 <.0001

CalYr 2011 1 0.2672 0.0219 0.2242 0.3101 148.81 <.0001

CalYr 2012 0 0.0000 0.0000 0.0000 0.0000 . .

Scale 0 2.0939 0.0000 2.0939 2.0939

Some other output that may be useful:

Class Level Information

Class Levels Values

CalYr 13 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009

2010 2011 2012

Criteria For Assessing Goodness Of Fit

Criterion DF Value Value/DF

Deviance 1040 4559.5819 4.3842

Scaled Deviance 1040 1040.0000 1.0000

Pearson Chi-Square 1040 4674.5441 4.4948

Scaled Pearson X2 1040 1066.2219 1.0252

Log Likelihood 157709.7869

Full Log Likelihood -5447.7431

AIC (smaller is better) 10927.4862

AICC (smaller is better) 10928.0098

BIC (smaller is better) 11006.8821

Re: ESTIMATE statement: age-adjusted rates using Poisson models in PROC GENMOD

SteveDenham — Tue, 08 Apr 2014 19:05:59 GMT

This approach will give you the estimate at age=0, agesq=0 and ageneg1=0. First off, that won't be consistent, if ageneg1 is defined as 1/age, so the value you end up predicting is meaningless. Why not shift to LSMEANS, rather than ESTIMATE? The values obtained would be at the mean values of age, agesq and ageneg. By careful use of the AT= option, you would get the best linear unbiased estimates (LSMEANS) at the specified values of age, agesq and ageneg.

If you intend to keep to the ESTIMATE statement, you will need to include plausible values of the continuous covariates in the ESTIMATE statement.

Steve Denham

topic ESTIMATE statement: age-adjusted rates using Poisson models in PROC GENMOD in SAS Procedures

ESTIMATE statement: age-adjusted rates using Poisson models in PROC GENMOD

Re: ESTIMATE statement: age-adjusted rates using Poisson models in PROC GENMOD